3D Vision & Generative AI

Digit3D: 3D Multimodal MNIST Benchmark

Transforming 2D MNIST into rich 3D geometries (watertight meshes, sparse SDFs, and 6D normal-oriented point clouds) with continuous flow matching generative pipelines.

Khoi DO
Conquer3D Research & Open-Source Geometry Engine
Code Conquer3D Engine Overview Unconditional Generation Class-Conditional Generation Image-Conditional Generation Arbitrary-Resolution Generation Sparse Voxel Reconstruction Image-Conditional Mesh Generation PointCloud-Conditional Mesh Generation Limitations
Multi-Scale Geometry

Arbitrary-Resolution 3D Point Cloud Generation

Continuous 3D point cloud synthesis across arbitrary point counts from a single 2D input image via Mean Flow. Select a resolution below to dynamically inspect geometries or compare all scales side-by-side.

Point Density:
0 Digit 0 (Sample #004) N = 1024 pts
2D Input Digit 0 2D Input
1 Digit 1 (Sample #000) N = 1024 pts
2D Input Digit 1 2D Input
2 Digit 2 (Sample #002) N = 1024 pts
2D Input Digit 2 2D Input
3 Digit 3 (Sample #023) N = 1024 pts
2D Input Digit 3 2D Input
4 Digit 4 (Sample #005) N = 1024 pts
2D Input Digit 4 2D Input
5 Digit 5 (Sample #008) N = 1024 pts
2D Input Digit 5 2D Input
6 Digit 6 (Sample #001) N = 1024 pts
2D Input Digit 6 2D Input
7 Digit 7 (Sample #029) N = 1024 pts
2D Input Digit 7 2D Input
8 Digit 8 (Sample #021) N = 1024 pts
2D Input Digit 8 2D Input
9 Digit 9 (Sample #016) N = 1024 pts
2D Input Digit 9 2D Input

Multi-Scale Point Cloud Synthesis: Switch resolution tabs above to dynamically inspect densities or compare all scales side-by-side.